If by reverse scoring you mean newX

I agree with Daniel Wright that this a linear transformation, and go further to say that this will not affect any statistics in a linear model. The two variables will be correlated -1.0 so you should see shifts in the signs in the relevant portions of SEM, but no difference in any of the fit statistics.

FYI this implies that your advisor does not know what they are doing, but rather than pointing this out, I would politely accept their advice, reverse the scoring, and move on.

+1 for David L Morgan suggesting politeness!

Samaher Saud, I am confused by some of the text in your post, namely: "the software continues to provide the same results, and some tests were affected by this change". That seems to be contradictory.

Could you provide some clarification / expansion, please?

Apart from that, my impression is that most researchers and consumers of research regard greater amounts of agreement to be associated with higher numbers in questionnaires and analyses. Although you'd not done that when collecting your data, that's what your supervisor is recommending for your analyses (and, presumably presentation of your results) - for what might be good reasons.

Please also be aware that the option of "neutral" could mean several things, including "I agree sometimes, but disagree at other times", I don't have any opinion concerning the matter", "I don't understand the question", and "I couldn't care less". Chyung et al. (2017) have referred to a midpoint such as “neutral” to be a “dumping ground” for a range of disparate meanings that respondents intend to convey. Despite this, the "neutral" option is pervasive - and I'm not sure how you want to interpret your participants' use of it.

Here's the reference for Chyung et al.:

Chyung, S.Y., Roberts, K., Swanson, I., Hankinson, A.: Evidence-based survey design: the use of a midpoint on the Likert scale. Perform. Improv. 56(10), 15–23 (2017). https://doi.org/10.1002/pfi.21727

All the best with your research!

Robert Trevethan Thank you for your response. My survey aims to assess self-regulatory capacity in vocabulary, utilizing the framework proposed by Tseng et al. (2006). I am investigating differences among socio-demographic variables such as age, gender, level, and major. When I mentioned that the tests consistently yield the same results, I was specifically referring to the one-way ANOVA, which continues to indicate, for instance, that males exhibit greater metacognitive control than females. However, in reality, females actually demonstrate higher control. I am using Jamovi, and after reverse-coding the items and saving their mean scores, attempts to rerun the correlation analysis have been unsuccessful. Additionally, my SEM has ceased to provide results. I already completed all the analyses with the original Likert scale that began with "1) strongly agree." I was also advised to use MANCOVA instead.

I appreciate the reference you provided; I wanted to have a whole picture of the extent to which undergraduates are self-regulated in their vocabulary learning. If their responses are closer to 1, it suggests a higher level of control over their learning. However, my advisors recommend assessing it through group differences rather than this direct measurement approach.

Daniel Wright Thank you for your response. I have already collected my data with the Likert scale presented as 1) strongly agree, 2) agree, 3) neutral, 4) disagree, 5) strongly disagree.

I have completed the entire analysis, but when I reached the One-way ANOVA to compare differences, for example, among majors, science students with mean scores closer to 1 are expected to have more commitment control. My advisor suggested reverse coding all the items and rerunning the analyses to improve the readability of the results for the examiners. When attempting this (I am unsure if this is the linear transformation you mentioned), I encountered issues with the software 'Jamovi.' I am contemplating whether to keep the original direction as it is and simply clarify the Likert scale direction in the thesis. If you have any suggestions, I would appreciate them

David L Morgan Thank you for your answer. Is it common in research to analyze quantitative data with a Likert scale in this direction? Will it cause difficulty in the readability of these results by examiners?

Samaher Saud, thank you for trying to clear up my uncertainties, but I'm afraid I'm more confused than before. Perhaps the kind of problem(s) you're facing can be addressed only through familiarity with the exact nature of your data and the empirical background to your topic. And those things are probably very difficult to address satisfactorily in a succession of posts on RG.

So, I think it's best that I bow out and - again - wish you the best with your research.

I think it's worth mentioning that your revised response alternatives are superior to the originals. Unlike, the latter, they're symmetrical atound the implicit centrepoint (which, in your revision is explicit: "Neutral" or alternatively "neither agree nor disagree"). Your revision is also clearer in that I'm not sure I buy the directional difference of the original from Partly Agree to Slightly Agree.

One point to consider is that, given. a 5-point Likert scale like yours, there are those who would advocate polychoric correlations in place of Pearson correlations.

It does not seem to be a transformation as such. What happened to 6? You had 6 categories (1) Strongly Agree, 2) Agree, 3) Partly Agree, 4) Slightly Agree, 5) Disagree, 6) Strongly Disagree)) and after the transformation you have 5 (1) Strongly Agree, 2) Agree, 3) Neutral, 4) Disagree, 5) Strongly Disagree)? That might have very strange effects and could explain the discrepancy between 'the software continues to provide the same results, and some tests were affected by this change'. Please explain what you did exactly. Otherwise it would be a linear transformation and all results should have been the same.

On my knowledge, you have two mistakes: in 1) changing the direction and 2) scoring process. The original direction of item scores has been designed reversely, and the reduction of response category from six to five should not be done. Sometimes scyle developers remove the central or neutral category (e.g., neither agree nor disagree) and select the even number of response options in order to make an optimal number of options and also to increase the probability of choosing adjacent categories in camparison to probability of choosing a central category (i.e., to increyse the reliability and validity of the instrument).

It’s quite common to use reverse coding in survey research, and it’s generally acceptable to use a scale in this direction in a thesis. The key is to clearly explain your methodology in your thesis, including why you chose to reverse code and how you did it.

However, it’s important to note that changing the direction of the scale (i.e., reverse coding) should not change the results of your statistical tests, assuming the reverse coding was done correctly. If some tests were affected by this change, it might be worth double-checking your reverse coding process.

Here are a few things to consider:

One problem that frequently arose with students doing reverse-coding and scale reliability exercises in my classes on SPSS is that they often inadvertently included the first variable which was unrelated to the scale, typically a subject ID number. Needless to say, this made their results horribly inaccurate; however, the problem was easily corrected by removing the errantt variable from the analysis.

This may or may not apply to your problem.

As promised, files to perform several types of scale analyses in SPSS when R is incorporated. I also add an R file that will almost do the same thing. It might help to shape your ideas. I did not translate Dutch into English, but Google Translate will help you. I did not include many notes as it was a test to see if it would work. Some analyses may not work anymore if procedure from SPSS or R have been modifiied.

Fayçal Ferahi Tests of scales that included reversed items have concluded that this is not a wise practice because respondents are too likely to miss the reversal, and thus give a completely incorrect response.

Formally, no matter do one use reversed scales or not in presenting items to respondents, especially if you providing answers without numbering, just: Strongly Agree, Agree, Neutral, Disagree, Strongly Disagree. You may just further extract raw values for each item from 6.

One could imagine that for some cultures more common to assess something from strongly disagree to strongly agree, i.e. as:

Strongly Disagree, Disagree, Neutral, Agree, Strongly Agree.

So some respondents may "forget" how to assess items in google-forms for big questionnaires and make some "technical mistakes".

But... I'm not sure that exists some researches, that could prove or disprove it.